CertLakeDatabricks Cert Prep

Study guidePro

Databricks Generative AI Engineer Associate

Every domain of the Databricks Generative AI Engineer Associate exam explained, with decision tables, runnable Python and SQL, diagrams and the traps the questions set, following the March 2026 exam guide.

Aligned to the March 2026 exam guide · checked against the Databricks docs on 27 October 2026 · certlake.com/exam/genai-engineer-associate/study-guide

Exam at a glance

Questions45 scored, multiple choice and multiple select, plus up to 10 unscored
Time90 minutes
Exam languagesEnglish, Japanese, Brazilian Portuguese, Korean
Exam guideExam published 18 March 2026
DomainWeight≈ Questions
Design Applications14%6
Data Preparation14%6
Application Development30%13–14
Assembling and Deploying Applications22%10
Governance8%3–4
Evaluation and Monitoring12%5–6

1. Design Applications14% of the exam

From business goal to pipeline

Start from what goes in and what must come out, then pick the components between them.

RequirementInputs → outputsComponents
Answer from internal documentsQuestion → grounded answer with citationsRetriever over an index, prompt template, LLM
Route support emailsEmail text → team labelClassification (a small model, or an LLM with few-shot examples)
Shorten a memo field to one lineParagraph → one sentenceSummarization
Contracts to a tablePDF → fields (party, date, term)Extraction with a fixed output schema
Answer "is this above target?"Question → number and verdictAgent with a data-lookup tool, then a compare step

A compound AI system combines several components (retrievers, models, tools, code) instead of one model call. A chain runs fixed steps in a fixed order; an agent lets the LLM decide which tool to call next, and when to stop. Evaluation-driven development: define the evaluation set and metrics before building, then keep a change only if the scores improve.

Prompts for a fixed format

  • Put the format rules in the system prompt: "Respond only with JSON with keys summary and sentiment."
  • Add a one-shot or few-shot example of the exact output. Few-shot is also the quickest way to start a task with only a handful of labelled examples.
  • Delimit the input (<ticket>…</ticket>) so instructions and data don't mix.
  • When code parses the reply, ask the API for structured output (response_format with a JSON schema) rather than relying on the prompt alone.

Tools for multi-stage reasoning

Define each tool to do one thing, with a clear name and description the LLM can choose from, and order them so each step's output feeds the next: look up sales, then look up the target, then compare. Gather knowledge first (retrieval, SQL), then act (write, send). Pass structured fields, not prose, between stages.

Agent loop: the user request goes to the LLM, which either calls a tool or answers; each tool result is added to the context and the LLM decides again, until it gives the final answerUser requestplus system prompt and toolsLLM decidestool call or answerTool runsUC function, indexresult added to the contextFinal answerno more tools neededEach step is a spanin the MLflow trace
A chain fixes the steps in advance; an agent's LLM chooses the next tool on each turn of this loop.

Agent Bricks

Agent Bricks builds and optimises a governed agent from a description and your data, without hand-building retrieval and serving. The exam names these three types; the docs now list Knowledge Assistant and Supervisor Agent as legacy.

NeedAgent Bricks type
Q&A chatbot with citations over documents in a UC volume or indexKnowledge Assistant
One entry point that routes to several agents, Genie spaces, UC functions or MCP serversMultiagent Supervisor (Supervisor Agent)
Unstructured documents → structured fields in a tableInformation Extraction

The rest of this guide is part of Pro

5 more domains, explained with snippets, decision tables and diagrams:

  • 2. Data Preparation · 14%
  • 3. Application Development · 30%
  • 4. Assembling and Deploying Applications · 22%
  • 5. Governance · 8%
  • 6. Evaluation and Monitoring · 12%
See Pro plansRead a full guide free (Data Engineer Associate)

Know it, or just recognise it?

The free diagnostic shows which domains you'd lose marks on, in about 10 minutes and without signing up.

Take the free diagnostic

Cheat sheet

The two-page cheat sheet: rules, defaults and traps for the last week before the exam.

Open the cheat sheet

Spotted something out of date? Email support@certlake.com. CertLake is not affiliated with Databricks, Inc.